chore(voice): default NOTIFY_VOICE_STYLE to max

[ajlennon]

Summary

• Default MCQ/notification TTS style is now max (was charlie-t), matching cursor-config NOTIFY_VOICE_STYLE.
• Docs (SETUP.md, VOICE-BACKENDS.md) and doctor hint updated.
• Still non-Alex: boundary rejects neutral; override via NOTIFY_VOICE_STYLE remains.

Hosts / platforms touched

• Cursor (Linux Gtk)
• Claude Code
• Claude Desktop / other stdio host
• Windows Phase 1 (tkinter text-only)
• Docs only (no behaviour change)

Test plan

CI must stay green (test, secrets-hygiene). Also say what you ran:

Check	Result	Notes
uv run python scripts/test_match_transcript.py	Skip	no transcript change
uv run python scripts/test_doctor.py	Skip	string-only doctor hint
uv run python scripts/test_acks.py	Skip	default style env; pool regen not in this PR
uv run python scripts/test_install.py	Skip
uv run python scripts/test_dialog_ergonomics.py	Skip
uv run python scripts/test_windows_backend.py	Skip
uv run python scripts/test_contracts.py	Skip
uv run python scripts/test_mcq_images.py	Skip
check_setup in a real host	Skip
ask_multiple_choice dialog click	Skip	needs Cursor Reload after mcp.json env
Image preview in Gtk dialog (image/images)	Skip
Keyboard 1–8 / Enter / Esc	Skip
Freeform / Something else	Skip
Speak / duck (if claiming voice)	Pass	TTS /tts with style=max verified; YubiKey prompt WAV regenerated locally

Environment: Linux Framework laptop; Qwen3-TTS on ai-tools :8200; Max style installed.

Behaviour contract

• No removal/rename of MCP tools without a migration note (ask_multiple_choice, check_setup, setup_guide, record_platform_feedback)
• Cursor absolute-uv docs kept (GUI thin PATH)
• No lab IPs / home paths / secrets in tracked files
• If adding a host: update Tested platforms table (Verified vs Not yet reported must match your results)

AI assistance

• Human-authored
• AI-assisted (verified TTS health lists max, regenerated YubiKey WAV with style check, defaults/docs only in this PR)